Typical data centers do not provide a pool of byte-addressable memory (e.g., RAM, non-volatile memory), to be accessed by other compute devices on an as requested basis. Rather, such memory is installed into each separate compute device, which can lead to over-provisioning of local memory in each compute device if the compute devices typically do not use all of the memory that they are equipped with. Further, in systems that provide access to remotely accessible resources, the condition (e.g., health) of the underlying hardware of those resources is often unknown to the compute device that is utilizing the resource. As such, the compute device may only be notified of an issue with the condition of the remote hardware when the hardware fails completely, potentially resulting in loss of data.